# How the Nordics Are Building with AI - Featuring Legora, Corti & Root Bio | Build-a-Claw Stockholm

## Executive summary

This panel explores the deployment of agentic AI in highly regulated and complex industries, including healthcare, legal tech, and biotech. Key insights focus on moving AI beyond simple question-answering to performing end-to-end, meaningful tasks. Speakers highlighted the necessity of building sovereign, robust stacks with programmatic guardrails, the shift from bits to atoms in scientific discovery, and the strategic importance of a customer-first, fast-iteration development cycle, even when dealing with mission-critical systems.

## Key takeaways

- Agentic AI in Regulated Environments: In healthcare, agents are being deployed in mission-critical environments, such as the biggest EHR system in Europe. Success requires programmatic guardrails and strict checks between sub-agents, rather than relying solely on prompt-based guardrails. (Timestamp: [Lars's section])
- The Future of Legal Work: AI will automate the 'grunt work' of legal practice (e.g., research, drafting, contract review), allowing human lawyers to focus on high-value tasks requiring human judgment, such as risk appetite and client understanding. This process aims to democratize legal advice. (Timestamp: [Jacob's section])
- AI in Biological Discovery: The process is evolving from 'bits to atoms,' allowing users to move from a prompt to a molecule in the lab. AI can design completely new molecules, which are then tested using automated robotic systems. (Timestamp: [Martin's section])
- Development Strategy: Speed vs. Robustness: While initial development should prioritize speed and rapid iteration (even if it means accruing tech debt) to achieve product-market fit, the final product must be built robustly and reliably. (Timestamp: [Jacob's section])

## Technical details

- Agentic Infrastructure: Corti provides an agentic framework that services LLMs, speech-to-text, text generation, and deep workflows, enabling developers to build applications across multiple sectors (healthcare, life science, finance). (Timestamp: [Lars's section])
- Sovereign AI Stacks: Due to regulatory concerns (e.g., EU regulations), building a fully sovereign stack is necessary. This involves deploying the entire stack on any cloud that provides the necessary GPU resources, favoring open models for flexibility. (Timestamp: [Lars's section])
- Model Selection Criteria: When selecting AI models, the criteria must balance price, performance, cost, and latency with crucial governance and risk profile considerations. (Timestamp: [Jacob's section])
- Open-Source Development: The use of open-source environments (like Kommodity) and open models is encouraged to demonstrate full-stack transparency and cloud-agnostic deployment. (Timestamp: [Lars's section])

## Practical implications

- For builders, prioritize a customer-first approach (distribution first) over building a cool product in isolation.
- When designing AI systems for regulated industries, implement programmatic guardrails and strict checks between agents, rather than relying solely on prompt engineering.
- Be prepared for rapid technological change; the 'form factor of the frontier of AI changes every six plus twelve months.'
- Focus on building robust, sovereign stacks that can operate across different cloud environments to mitigate geopolitical risks.

## Topics

Agentic AI, Build Engineering, Artificial Intelligence, Biotechnology, Legal Technology, Sovereign Computing, docs.corti.ai, corti.ai/safety, Stockholm AI Lab

Source: https://www.youtube.com/watch?v=LHJoYg17SuE